Tacrolimus Exposure Prediction Using Machine Learning

Abstract

The aim of this work is to estimate the area‐under the blood concentration curve of tacrolimus (TAC) following b.i.d. or q.d. dosing in organ transplant patients, using Xgboost machine learning (ML) models. A total of 4,997 and 1,452 TAC interdose area under the curves (AUCs) from patients on b.i.d. and q.d. TAC, sent to our Immunosuppressant Bayesian Dose Adjustment expert system (http://www.pharmaco.chu-limoges.fr/) for AUC estimation and dose recommendation based on TAC concentrations measured at least at 3 sampling times (predose, ~ 1 and 3 hours after dosing) were used to develop 4 ML models based on 2 or 3 concentrations. For each model, data splitting was performed to obtain a training set (75%) and a test set (25%). The Xgboost models in the training set with the lowest root mean square error (RMSE) in a 10‐fold cross‐validation experiment were evaluated in the test set and in 6 independent full‐pharmacokinetic (PK) datasets from renal, liver, and heart transplant patients. ML models based on two or three concentrations, differences between these concentrations, relative deviations from theoretical times of sampling, and four covariates (dose, type of transplantation, age, and time between transplantation and sampling) yielded excellent AUC estimation performance in the test datasets (relative bias < 5% and relative RMSE < 10%) and better performance than maximum a posteriori Bayesian estimation in the six independent full‐PK datasets. The Xgboost ML models described allow accurate estimation of TAC interdose AUC and can be used for routine TAC exposure estimation and dose adjustment. They will soon be implemented in a dedicated web interface.

A hybrid machine learning/pharmacokinetic approach outperforms maximum a posteriori Bayesian estimation by selectively flattening model priors

“…Applications of ML to MIPD to date have found that ML models are often able to accurately estimate past drug exposure 24,25 , predict future drug exposure 26-28 or select doses [29][30][31][32] . However, the improvement in accuracy from these earlier approaches comes at the expense of pharmacological interpretability and the ability to simulate patient response to alternative dosing regimens 24,33,34 . An advantage of the combination of ML and PK models as described here is that clinical decision making is augmented by ML while maintaining the ability to forecast patient PK and extract mechanistic insight from PK parameter estimates.…”

Machine Learning as a Novel Method to Support Therapeutic Drug Management and Precision Dosing

“…10 The paper is a follow-up to a recent publication in the journal by the same group, on machine learning for the precision dosing of tacrolimus. 11 In both studies they used XGBoost, which stands for "Extreme Gradient Boosting, " an open-source software library with which large data sets are fit into a mathematical model, while maintaining high computational speed. Interestingly, the machine learning approach outperformed the PK model-informed method for the indications and times after transplant tested, and the authors are planning to implement the machine learning models as an expert system in the ISBA service.…”

Mycophenolic Acid Exposure Prediction Using Machine Learning

“…[6][7][8][9][10][11] Machine learning (ML) is widely used in numerous applications, including pharmacology (PubMed Entry "Machine learning" & "Pharmacology" = 600 in 2018, 846 in 2019) especially in structure/activity predictions 12,13 or drug discovery 14 but only a few applications to predict drug exposure, PK parameters, or optimal dose exist. [15][16][17][18][19] Recently, we successfully applied a ML approach for tacrolimus AUC estimation that yielded better performance in terms of relative bias or imprecision vs. reference trapezoidal rule AUC than MAP-BE, even with only two samples. 19 We used extreme gradient boosting (Xgboost R package) where simple regression trees are iteratively built by finding among all the input variables, split values that minimize the prediction error.…”

“…[15][16][17][18][19] Recently, we successfully applied a ML approach for tacrolimus AUC estimation that yielded better performance in terms of relative bias or imprecision vs. reference trapezoidal rule AUC than MAP-BE, even with only two samples. 19 We used extreme gradient boosting (Xgboost R package) where simple regression trees are iteratively built by finding among all the input variables, split values that minimize the prediction error. The iterative process constructs an additional regression tree of the same structure that minimizes the residual errors of the previous regression tree.…”

“…There are some differences between this study and our previous one with tacrolimus. 19 Indeed, tacrolimus PopPK is not easy to model, but it has obvious associations with covariates that explain part of the interindividual variability, the dose-exposure relationship is good, and the AUC is fairly correlated with single timepoints (the first being the trough level, used in routine practice). 30 All these elements may explain why our first ML algorithms were successful.…”